Fault tolerant feedforward neural network with median neuron input function
Andrzej Rusiecki · Electronics Letters · 2005
A simple and effective method to achieve increase of fault tolerance of feedforward neural networks (FFNs) is proposed. Results of implementation and simulation of nets with the new median neuron input function, and the dedicated learning algorithm, are presented and compared with traditional FFNs. The better performance of the novel approach in the presence of faults is shown.